Into The Mind of AI Guru Gary Marcus: On The Limits of AI

CityAge interviewed Gary Marcus in early November:

There’s all this hype about how AI
can replace everybody’s job and so forth, but it can’t actually.

The leading technique right now is deep learning
and it reminds me of Winston Churchill’s line [about democracy] because it’s not very good, but it’s better than the rest of what we’ve got. But it still has a lot of serious problems. 

Robots are mostly creatures of assembly lines right now,
where things are extremely well controlled with very good cameras. People don’t really — or researchers don’t — know how to get robots to manipulate objects freely in somebody’s home.

AI  can often get answers that are approximately correct to some degree, but you can’t really count on it.
So we are finding, for example, that people can make driverless cars that work some of the time, but they don’t work all at the time. People can make chatbots that understand maybe 80% of what you say, but not a hundred percent of what you say. 

For some purposes that’s good enough.
If you’re recommending an advertisement and you’re correct four times out of five, that might be fine. But, if you’re building a driverless car or you’re trying to build a robot that works in your home, that level of accuracy isn’t anywhere near good enough. 

My favorite example of when it went wrong is (with) a well known China entrepreneur
whose picture was on the side of a bus and got a traffic ticket for jaywalking. The surveillance system was looking for jaywalkers and didn’t understand that a picture on the side of a bus wasn’t the same as a real person.

Geoffrey Hinton from Toronto talked about how radiologists were all going to be replaced
in five years. Now it looks like AI will help radiologists, but it’s not going to replace them. It’s not anywhere near that level right now. So there have been a lot of things that have been promised and really not delivered.

My general sense so far is that quantum computing
works only on some very specialized problems. It’s not going to solve AI, but it might be part of the picture.

[Gary Marcus is
a scientist, best-selling author, and entrepreneur. He is Founder and CEO of Robust.AI, and was Founder and CEO of Geometric Intelligence, a machine learning company acquired by Uber in 2016. His latest book is Rebooting AI: Building Machines We Can Trust]

Contact Us

We're not around right now. But you can send us an email and we'll get back to you, asap.

Not readable? Change text. captcha txt